The kernel outlives the laptop

You start the cross-validation run at 4:10pm. It'll take two hours. At 4:40 the train enters a tunnel and the laptop's Wi-Fi dies.

Nothing happens to the run. Jupyter is listening on port 8888 inside your container, which makes it live at …-http-8888.… with a certificate. The kernel is on the server's cores, not your battery. You reopen the URL on your phone in the next station, and the progress bar is at 23%.

At 6:15 a colleague asks to see the fold scores. You send her the URL. She's in the same kernel, same variables, same plots. No tunnel, no jupyter notebook --ip=0.0.0.0 warnings, no port forwarding over SSH.

You close the laptop for the night. The kernel stays warm with the 3GB DataFrame loaded, ready for tomorrow.